Web Services Mash-up : Flickr Cal Henderson O’Reilly Emerging Technology Conference March 14-17, 2005

What’s Flickr? A website – flickr.comflickr.com A photo-sharing application The centre of a big distributed system An open set of APIs –flickr.com/services/flickr.com/services/

Web services? What the heck are web services? The future of the Interwebnet!!!1 Web 2.0 An ‘emerging technology’ –(no, really)

But seriously? Probably a service accessed over the web RSS/Atom/RDF Feeds The Flickr API (Also, Flickr-specific junk)

Logical Architecture Page Logic Business/Application Logic Database Photo Storage API Logic EndpointsTemplates Users 3 rd Party Apps Flickr Apps Node Service Flickr.com Parser

Logical Architecture Page Logic Business/Application Logic Database Photo Storage API Logic EndpointsTemplates Users 3 rd Party Apps Flickr Apps Node Service Flickr.com Parser

API Architecture Endpoints Users Applications HTTP Transport Net / Local Transport

Transport Voodoo Like any decent Internet ‘standard’, there’s more than one A quick guide to the trendy ones…

SOAP Simple Object Access Protocol Now just SOAP –(not so simple anymore) Google uses it

SOAP Response <s:Envelope xmlns:s= xmlns:xsi= xmlns:xsd= > [escaped-xml-payload]

XML-RPC XML Remote Procedure Call Used by the blogging APIs Created by Dave Winer in 1998 –Because SOAP was taking a long time

XML-RPC Response [escaped-xml-payload]

REST Representational State Transfer –Crazy name Thanks Roy Fielding at Apache It’s really simple –Just XML over HTTP –(Though purists say it’s only HTTP GET)

REST Response [xml-payload]

Page Scraping Been around for ever HTML-over-HTTP Volatile interface Makes site owners angry

Scrumjax Probably the future

Offering Web Services Be transport agnostic –Some people love SOAP, some love REST –Make them all (somewhat) happy Beware of ‘shitty coders’

Performance Problems People can scrape your site and pull a lot of pages in a short time This is bad But API abuse (even accidental) can be a lot worse

An example Someone writes a trendy screensaver app for Flickr which shows recent photos. It checks for new photos every 2 seconds A bunch of people download it

Danger! With 100 users, that’s 50 hits per second If it’s making a particularly taxing database call, it’s going to cause problems

Possible solutions Incorporate caching into API bindings Enforced policy –Through API keys, etc. Cache at the host application level Monitor things closely